2

I am looking for datasets that would list the acceptance rates of graduate school applicants based on their undergrad schools. I.e. X % of students from undergraduate school Y applying to graduate school Z are admitted. Ideally, the dataset should also indicate the number of applicants.

I am mostly interested in graduate schools based in the United States, and computer science departments.

Improve this question
7
  • 9
    I am reasonably confident that an open dataset like this doesn't exist. First, because graduate admissions is a secretive process in the US, and second, because the combination of "attended Y and applied to Z" would be considered PII and shouldn't be released for privacy reasons. If you explain why you want this data (e.g., are you trying to measure elitism in graduate school admissions?) it might help someone guide you to another kind of dataset that might suit the same purpose.
    – ff524
    Commented Jan 6, 2016 at 19:21
  • 1
    @ff524 Purpose: curiosity. These statistics are aggregates, so there should not any PII issue (uneducated guess). One can of course exclude cases where a very low number of students from Y applied to Z, if needs be: the percentage wouldn't be much meaningful anyway.
    – Franck Dernoncourt
    Commented Jan 6, 2016 at 19:28
  • 3
    At least in my department, to avoid releasing PII you would have to exclude almost all the data. For most undergrad schools, the number of students applying is nowhere near the number you would need in order to avoid privacy implications. In some cases, even if the number is large, those schools would still need to be excluded because releasing it would enable others to guess whether someone they already know attended Y and applied to Z was accepted or rejected.
    – ff524
    Commented Jan 6, 2016 at 19:34
  • 1
    @ff524 What number does one need in order to avoid privacy implications in the US? I'd guess one could skip years, or perturb the data in some other ways (e.g. random noise). Otherwise data may come from applicants themselves, e.g. thegradcafe.com (which claims to have "374390 grad school admission results in the database".)
    – Franck Dernoncourt
    Commented Jan 6, 2016 at 19:38
  • 4
    The gradcafe data is so biased (sampling bias, reporting bias), I wouldn't consider it anywhere close to a reliable indicator of "grad admission acceptance rate based on undergrad school."
    – ff524
    Commented Jan 6, 2016 at 19:41

1 Answer 1

Reset to default
4

The specific dataset you seek is likely unavailable because of concerns about personal identifying information and the confidentiality of the process. The number of applicants from small colleges or to small CS programs in any given year is in many cases too low for effective anonymity.

Many good undergrad programs will tell you about where their graduates went in terms that they are comfortable with, e.g. "x% went to grad school;" some will even list exactly which graduates went where - usually with the permission of both the graduate and the department head. I don't think anybody's gathered this into a single dataset.

However, certain aspects of your curiosity may be satisfied by graphs and compiled data from a step later down the line, about which universities hire grads from which other universities for their CS faculty. That kind of information can be found here (compiled by Jeff Huang and his students at Brown University, analyzed by Jürgen Pfeffer at Carnegie Mellon):

(source: pfeffer.at)

Improve this answer
4
  • Thanks, how about large colleges/CS programs, and perturbing the data (e.g., random noise or skip years)?
    – Franck Dernoncourt
    Commented Jan 6, 2016 at 20:16
  • The key issue is lack of motivation/incentives to gather such a data set: for what purpose would it be used? There are so few large colleges/CS programs that limiting to those wouldn't tell you much, and analyzing random noise is not likely to be helpful in answering a motivating question. Data could be aggregated over years (skipping doesn't reduce the PII specificity issue) at the cost of current/predictive value (programs change over time). Data collection should be motivated by a specific question, so as to guide decision-making between the trade-offs involved.
    – WBT
    Commented Jan 6, 2016 at 20:22
  • 1
    This is exactly the link that came to mind when I read the question, but I was having trouble recalling where I read it. Thanks for posting it! I also recall something like a network analysis of "longest connection length", with the general idea that "for top institutions, it really is a small world" - that at top institutions they had 1-link long connections to everywhere, while at smaller institutions it might take 4+ links to get some other department. The one's I'm thinking of don't answer the OPs questions either though, I'm afraid.
    – BrianH
    Commented Jan 6, 2016 at 22:14
  • This is interesting data. It's different from what the OP requested in two ways: besides for being about a different stage (as acknowledged), it only includes successful applicants, not rejected applicants. i.e. assuming these 51 school are the only schools in existence, you can use it to say "N% of hires at Z come from Y" or "M% of Y graduates who get positions end up at Z." But you can't say "N% of applicants to Z from Y are hired," which would be analogous to what OP wanted for grad admissions ("X % of students from undergraduate school Y applying to graduate school Z are admitted").
    – ff524
    Commented Jan 6, 2016 at 23:08

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .